Open-Source AI

Openly available data, code, and infrastructure have been critical to the advancement of science, technological innovation, economic growth, and democratic governance. These open resources have been built and shared in the context of commitments to open science, to expanding industry and markets, and to the principle that some technologies should be widely available for maximum public benefit, while allowing for control of access to data, code, and infrastructure as necessary for safety and security purposes. As AI governance is developing globally, policy solutions related to people's rights and safety have tended to focus on increasing transparency and accessibility of AI systems to improve accountability of AI developers and deployers to the public. But beyond that, a robust open foundation model ecosystem is crucial to enabling a diverse, innovative, and competitive environment for technological innovation, as well as to addressing concerns about any single entity accumulating excessive sector influence. If well executed, an open foundation model ecosystem will expand collective understanding of AI beyond those who currently build and release models, spurring ingenuity and enabling a powerful base for innovation for researchers and developers across a range of sectors. The decision around whether model weights should be made widely available or not requires a full understanding of the threats and benefits associated with open foundation models. A more comprehensive and nuanced discussion of open versus closed foundation model access, including spectrum of access considerations, is essential.

Recommendations to the US Department of Commerce (NTIA) on Open Foundation AI Models, March 2024